Can you please post the output of

make VERBOSE=1 test_box2p

after performing a "make clean".


In any case, you have to use the AMGBackend as linear solver. This is the 
default, but in 2.12, the linear solver is set to ILU0BiCGSTABBackend in 
lensproblem.hh, which isn't parallel. So you have to replace the line

SET_TYPE_PROP(LensBoxProblem, LinearSolver, ILU0BiCGSTABBackend<TypeTag> );
by
SET_TYPE_PROP(LensBoxProblem, LinearSolver, AMGBackend<TypeTag> );
and to
#include <dumux/linear/amgbackend.hh>
in that file.

Anyway, that's not the problem that you encounter here. Apparently, MPI is 
detected properly, but you are not employing it.

Kind regards
Bernd



--
_______________________________________________________________

Bernd Flemisch                         phone: +49 711 685 69162
IWS, Universität Stuttgart             fax:   +49 711 685 60430
Pfaffenwaldring 61            email: [email protected]
D-70569 Stuttgart            url: www.hydrosys.uni-stuttgart.de
_______________________________________________________________
________________________________
Von: lc <[email protected]>
Gesendet: Mittwoch, 30. Januar 2019 13:19:47
An: Flemisch, Bernd; DuMuX User Forum; Koch, Timo
Betreff: Re: AW: [DuMuX] set_singular_limit and parallel execution


Here is the log,


Thank you,

Lorenzo


On 30.01.2019 14:53, Bernd Flemisch wrote:
Are you sure that dunecontrol finds MPI? Can you post the output of dunecontrol?

Bernd

On 01/30/2019 12:46 PM, lc wrote:

Good morning,


I updated the virtual machine as you may see from the lscpu_new file attached.

Then, I try to run both sequential 2p test_impesadaptive and 2p implicit 
test_box2p (input file attached) with 1 4 and 8 cores straight from the Dumux 
2.12 test suite without any other modification (just the grid in the input 
file).


I still get the same behaviour:

1 core -> 302 s

4 cores -> 337 s (output attached)

8 cores -> 470 s


It seems that as you noted, these processes do not communicate each other. Do 
you agree?


These testcases, should be fully parallel, so what's wrong?


Thank you,


P.S. I also have other pending questions ...


Kind regards,

Lorenzo



On 24.01.2019 14:42, Flemisch, Bernd wrote:

Timo is of course right. Nevertheless, your parallel output shows that the 
program isn't executed in a distributed way. It simply executes the full 
sequential code on four processes that don't communicate with each other. The 
four processes occupy two physical cores resulting in a degradation in 
performance. The sequential (in the sense of temporal discretization and 
coupling) 2p2c model is not parallel. The fully implicit ones are parallel, 
also the sequential 2p model is parallel. Let us know if you experience the 
same problems with a truly parallel model. If you do so, please consider Timo's 
comment that you choose a number of processes that is less or equal than the 
number of physical cores on your machine.


Kind regards

Bernd


--
_______________________________________________________________

Bernd Flemisch                         phone: +49 711 685 69162
IWS, Universität Stuttgart             fax:   +49 711 685 60430
Pfaffenwaldring 61            email: 
[email protected]<mailto:[email protected]>
D-70569 Stuttgart            url: 
www.hydrosys.uni-stuttgart.de<http://www.hydrosys.uni-stuttgart.de>
_______________________________________________________________
________________________________
Von: Dumux 
<[email protected]><mailto:[email protected]>
 im Auftrag von Timo Koch 
<[email protected]><mailto:[email protected]>
Gesendet: Donnerstag, 24. Januar 2019 11:01:56
An: DuMuX User Forum; lc
Betreff: Re: [DuMuX] set_singular_limit and parallel execution


Hi Lorenzo,

you can't scale well to 4 cores because your CPU only has two cores. The two 
additional ones are only hyperthreads. Please repeat the same numerical 
experiment on a machine with more cores.

Timo

On 24.01.19 10:37, lc wrote:

On 18.01.2019 21:52, Flemisch, Bernd wrote:
Can you please post the complete output for the run with 4 cores? Plus the 
initial vtk files? And, pretty please with sugar on top, the number of physical 
cores on your machine?


Good morning,

I enclose the requested files for the sequential (1 processor) and parallel (4 
processors) simulations plus some detail on my machine.

Thanks for help,

Lorenzo




_______________________________________________
Dumux mailing list
[email protected]<mailto:[email protected]>
https://listserv.uni-stuttgart.de/mailman/listinfo/dumux


--
_______________________________________________________________

Timo Koch                              phone: +49 711 685 64676
IWS, Universität Stuttgart             fax:   +49 711 685 60430
Pfaffenwaldring 61        email: 
[email protected]<mailto:[email protected]>
D-70569 Stuttgart            url: 
www.hydrosys.uni-stuttgart.de<http://www.hydrosys.uni-stuttgart.de>
_______________________________________________________________


--
_______________________________________________________________

Bernd Flemisch                         phone: +49 711 685 69162
IWS, Universität Stuttgart             fax:   +49 711 685 60430
Pfaffenwaldring 61            email: 
[email protected]<mailto:[email protected]>
D-70569 Stuttgart            url: 
www.hydrosys.uni-stuttgart.de<http://www.hydrosys.uni-stuttgart.de>
_______________________________________________________________

_______________________________________________
Dumux mailing list
[email protected]
https://listserv.uni-stuttgart.de/mailman/listinfo/dumux

Reply via email to